noise canceling in 1-d data: presentation #3
DESCRIPTION
Feb 2 nd , 2005 Size Estimates and Floorplan. Noise Canceling in 1-D Data: Presentation #3. Seri Rahayu Abd Rauf Fatima Boujarwah Juan Chen Liyana Mohd Sharipp Arti Thumar. M2. Project Manager: Bobby Colyer. Overall Project Objective: - PowerPoint PPT PresentationTRANSCRIPT
Noise Canceling in 1-D Data: Presentation #3
Seri Rahayu Abd RaufFatima BoujarwahJuan ChenLiyana Mohd SharippArti Thumar
M2
Feb 2nd, 2005Size Estimates and Floorplan
Overall Project Objective: Implementing Noise Cancellation Algorithm in Hardware
Project Manager: Bobby Colyer
Status
• Design proposal (Done)• Architecture proposal (Done)
– Behavioral Verilog simulation and test bench (Done)
• Size Estimates and Floorplan:– Structural Verilog (85%)– Revised transistor count (Done)– Floorplan (Done)
• To be done:– Schematics (15%)– Layout (5%)– Spice simulation
FP
A
w0 RO
M
FP
M
cos
FP
A
sin
FP
M
x2
FP
S
out
input
FP
MShifter
FP
M
FP
M
x1
error
mu
sumw0
FP
AF
PA
w1w2
Last week’s Block Diagram…
Revised Block Diagram
counter ROM
FPMultiplier
1
FPMultiplier
2
0
1
0
1
FPMultiplier
3
FPAdder
2
FPAdder
1
FPAdder
3
FPSubtrat
or
Adder
FPSubtrat
or
Convert
Convert
sin
cos
Regx1
Regx2
X1 wireRegw 1
W1 wire
X2 wireRegw 2
W2 wire
RegDatu
m
Rege
Regmu
1000000000
Datum-offset
RegInput
RegOut
Alternator
RegOf f s
et
0
1
0
1
0
1
0
1
Behavioral Simulation Output
Converted Values from Verilog Simulation:Input: 9.3125 Output: 9.8227359 Error: -0.5
Original C Code Simulation Results:Input: 9.325461 Output: 9.650836 Error: -0.325
**Difference in values due to 16-bit precision
Results Comparisons (Output)
Result Comparison (Error)
Structural Verilog Example: Floating Point Multiplier
Metal Directionality
M2
M4
M1
M3
Note:Depending on the complexity of the blocks, metal 1 and metal 2 may be used in both directions.
Local vs Global Interconnects
M2
M4
M1
M3
Note:
Vdd! & Ground: M1, M2Internal Routing: M1, M2
Clock & Reset: M3, M4Global Routing: M3, M4
Last week’s transistor count…
Part Transistors
16-bit FPA 5x500 = 2500
16-bit FPM 5x4000 = 20000
SRAM 500
ROM 1600
Registers 7x16x14 = 1568
Total ≈ 26168
A more accurate transistor count…
Part Transistors
16-bit FPA 5x1700 = 8500
16-bit FPM 3x2028 = 6084
Registers 10x16x14 = 2240
ROM 800
Converter 2x312 = 624
MUX/DEMUX 384
Adder 248
Counter 214
Alternator 64
Total ≈ 19146 + Misc ≈ 22000
Tentative Floorplan
Total Area:
~190,000 µ²
Area EstimatesPart Area (µ²)
16-bit FPA 5x140x110 = 77000
16-bit FPM 3x140x106 = 44520
16-bit Register 10x33x24 = 7920
ROM 43x180 = 7740
MUX/DEMUX 6x6.8x57 = 2325.6
Converter 2x20x40 = 1600
Counter 1004.5
Adder 28x35 = 980
Alternator 1x48 = 48
Total ≈ 143138 µ² + Misc ≈ 190000 µ²
Ratio/Density
• Aspect Ratio:– 415µ x 450µ = 1:1.08
• Block Porosity:– 22000 transistors/190000µ2
– 0.12 transistors/µ2
Last week’s challenges…
• Timing issues– Need to reuse hardware (multipliers)
(reduced from 5 to 3)– Clock skew (needs to be addressed)– Pipelined architecture to increase speed
and throughput (not needed)
• SRAM implementation (not needed)
• ROM implementation (completed)
• Transistor count is too high (reduced through hardware reuse and block optimization)
This week’s challenges…
• Still working on finalizing out designs for the floating point adders and multipliers– Wallace tree multiplier vs Array
multiplier• Wallace implementation is
smaller because it reuses the intermediate products of the individual bits
– Leading zero counter for normalizing block
Questions?